CN106529428A - Underwater target recognition method based on deep learning - Google Patents
Underwater target recognition method based on deep learning Download PDFInfo
- Publication number
- CN106529428A CN106529428A CN201610929714.XA CN201610929714A CN106529428A CN 106529428 A CN106529428 A CN 106529428A CN 201610929714 A CN201610929714 A CN 201610929714A CN 106529428 A CN106529428 A CN 106529428A
- Authority
- CN
- China
- Prior art keywords
- training
- data
- network
- zca
- power spectrum
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
- G06F2218/16—Classification; Matching by matching signal segments
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to an underwater target recognition method based on deep learning, which mainly solves the problem that the existing target recognition system mainly relies on the shallow model to extract the artificial feature, resulting in the recognition accuracy is not high. The method comprises the following concrete steps that (1) power spectrum characteristics of original ship target radiated noise data are obtained in segment; (2) the power spectrum characteristics are divided into a training data set and a testing data set; (3) training data are subjected to ZCA whitening pretreatment; (4) a stack-type self-coding network is constructed; (5) a deep network node is subjected to fine tuning; and (6) testing data are subjected to classification and recognition. Sea test data and experiments show that the deep network is utilized to learn deep characteristics of ship target radiated noise for performing classification recognition, the recognition rate reaches over 94%, and the underwater target recognition effect is improved, which has an important real application prospect for future recognition and monitoring of underwater ships and marine organisms.
Description
Technical field
The invention belongs to field of underwater acoustic signal processing, is related to pattern recognition and artificial intelligence, particularly should by deep learning
For the feature extraction to method of underwater vessel radiated noise, can be used for passive sonar system to naval vessel and the long distance of other submarine targets
From identification.
Background technology
It is modern that the acoustical signal of radiation under water produced when being navigated by water using Ship Target carries out the automatic target detection of passive type
The important component part of naval vessels and intelligent underawater ordnance.However, on the one hand due to being related to military field, research institution of various countries is to each
Of a relatively high from achievement in research confidentiality, the visible effective means of identification of open source literature is less;On the other hand due to marine environment
Complexity and underwater acoustic channel particularity, show as:Underwater acoustic channel when space-variant, non-linear and non-Gaussian system, reverberation with
And Environmental Noise Influence etc., to efficiently extract from the Ocean acoustics signals under Low SNR and can characterize target classification
Substitutive characteristics, inherently one technical barrier.These factors cause the Classification and Identification of automatic target under water of passive type to become
But developed relatively slow recognized problem by numerous scholars and research worker concern.
Resolve how it is critical only that effectively from different perspectives to the original letter of target for passive target Study of recognition under water
Number it is analyzed and studies.In recent years, substantial amounts of related academic documents and patented technology have been emerged.Wherein, conventional means master
The following aspects to be concentrated on:Based on time domain plethysmographic signal structure zero crossing distribution, peak-to-peak amplitude distribution, wavelength difference distribution and
The target classification identification that the characteristic vectors such as wave train area distributions are carried out.Based on traditional Power estimation, super-resolution Power estimation and high-order
The target classification that the characteristic vectors such as Ship Target line spectrum, continuous spectrum, modulated spectra and spectrum shape that Power estimation technology is extracted are carried out
Identification.Based on nonlinear theories such as chaos, point shapes, it is original data space with time serieses, carries out phase space reconfiguration or divide
Number Brownian movement modeling, by the Classification and Identification for calculating connection dimension, Lyapunov indexes or Hurst indexes are carried out.Based on short
When Fourier transformation, Gabor transformation, wavelet transformation and quadratic form when-frequency division cloth includes Cohen alanysis, Wigner-
The target classification identification that the time and frequency domain characteristics that the time frequency analysis means such as Ville distributions and Choi-Williams distributions are extracted are carried out
Deng.
Above-mentioned target identification method mainly extracts clarification of objective using Heuristicses, and their performance is largely determined
Selection due to the degree of association that operator is presented for the degree of awareness and sample of sample, experiment parameter and model easily receives people's
Subjective impact, generalization are restricted.Meanwhile, above method is based primarily upon shallow Model and carries out feature extraction, it is difficult to reflect mesh
The most essential internal relation of mark, causes the target characteristic for only obtaining by shallow Model met under practical situation to naval vessel mesh
Target is effectively recognized.Depth model is taken to carry out the new direction that feature learning is Underwater Targets Recognition.
The content of the invention
Technical problem to be solved
It is an object of the invention to overcome the artificial limitation for extracting target characteristic, it is proposed that a kind of based on deep learning
Underwater targets recognition.Learn the implicit structure in Ship Target radiated noise time-frequency spectrum using sparse self-encoding encoder, will be dilute
Thin own coding is cascaded as stack own coding neutral net, and combination low-dimensional feature obtains the higher-dimension of target and represents, recycles softmax
Grader is classified to target.Due to carrying out unsupervised learning using depth network, it is not only avoided that artificial extraction feature
Limitation and subjectivity, moreover it is possible to the relation of signal spectrum " part overall " is represented in overall compact mode, it is special to target
Property has more essential portraying.Simultaneously as supervised learning is carried out using softmax graders, can further improve identification
Accuracy, reaches to true ship seakeeping probability more than 94%.
Technical scheme
A kind of Underwater targets recognition based on deep learning, it is characterised in that step is as follows:
Step 1:Power spectrum characteristic is asked for original Ship Target radiated noise data sectional:
Obtain original Ship Target radiated noise sequence, carry out segment processing, the segment data for obtaining be x (n), n=0,
1,2 ..., N-1, according to formulaPower Spectral Estimation is carried out to segment data, by adjacent 4 segment data
Power spectrum carry out averagely, obtaining final power spectrum characteristic result of calculation:
Wherein f=(f1,f2,f3,…,fM) represent corresponding discrete point in frequency;
Step 2:Power spectrum characteristic is divided into into training dataset and test data set:
By power spectrum characteristic P (f) according to 3:1 randomly selects, and is divided into training dataset and test data set, will train number
According to each frame power spectrum characteristic of collection as string, training matrix P is constituted:
Wherein, Pi(fj) represent that the jth of i-th training data ties up power spectrum characteristic;
Step 3:ZCA whitening pretreatments are carried out to training data;
Step 4:Construct and train stack autoencoder network;
Step 5:Fine setting fine-tune depth network model:
Using back-propagation algorithm, all layers of stack autoencoder network are regarded into a model, it is in each iteration, excellent
Change all weighted values in network, obtain final network node parameter;
Step 6:Classification and Identification is carried out to test data:
According to the variance of the characteristic mean of training dataset, pretreatment test data, using the data for having processed as stack
The input of autoencoder network, calculates final classification results.
ZCA whitening pretreatments carried out to training data comprise the following steps that described in step 3:
The first step:Training data is standardized, the average of kth dimensional feature is first calculated
Seek the standard deviation of kth dimensional featureCan according to the following formula normalized process after training
Data:
Second step:ZCA whitening pretreatments are carried out, the covariance matrix of sample is first calculatedSingular value is done to Σ
Characteristic vector is tried to achieve in decomposition, is discharged by leu license-master time order, is constituted matrix U=[u1,u2,…,un], by eigenvalue by diagonal
Arrangement, composition characteristic value diagonal matrix S can be calculated as follows final ZCA albefaction results:
Wherein, ε is used as regularization parameter.
Construction described in step 4 simultaneously trains comprising the following steps that for stack autoencoder network:
The first step:Construction stack autoencoder network, constructs stack autoencoder network using sparse self-encoding encoder, and stack is certainly
The hidden layers numbers of coding network are 3, and the interstitial content of hidden layer is respectively 100-100-50, nodes and the ZCA albefactions of input layer
Pretreated training data intrinsic dimensionality is mutually all 512, and finally, increasing by one layer of softmax grader is used for output of classifying, always
Classification number be 3, the nodes of whole network are 512-100-100-50-3;
Second step:Using the training data X after ZCA whitening pretreatmentsZCATraining stack autoencoder network, first, according to instruction
The method of the single sparse self-encoding encoder of white silk, using XZCAAs input, the node parameter W of first hidden layer is calculated(1,1)And b(1,1),
Concrete grammar is as follows:
Step 1) for the input data set X comprising L sampleZCA, define the overall cost function of sparse self-encoding encoder
Wherein, λ is weight attenuation parameter, is set to 3e-3, and β is the parameter for controlling openness penalty factor, and being set to 3, KL is
Relative entropy;
Step 2) by W(1,1)And b(1,1)Random initializtion is one be close to zero value, using L-BFGS iterative algorithms to generation
Valency function is optimized, and iterationses are set to 100, is met the optimal solution of requirement;
3rd step:Using the activation value of the first hidden layer as the input of the second hidden layer, the second hidden layer is calculated with same method
Node parameter W(2,1)And b(2,1), by that analogy, successively train, obtain the parameter of all hidden layers;
4th step:Last hidden layer is exported into the input as softmax graders, by L-BFGS iterative algorithm iteration
The optimal solution for obtaining softmax graders 100 times.
Beneficial effect
A kind of Underwater targets recognition based on deep learning proposed by the present invention, compared with prior art with following
Advantage:
First, compared with traditional feature extracting method, present invention, avoiding it is artificial extract feature, but using it is sparse from
Encoder learns to input sample, by the unsupervised learning ability of self-encoding encoder, automatically excavate target input sample it
Between internal relation, so as to learn more effective feature.
Second, sparse self-encoding encoder is stacked by the present invention, constructs stack autoencoder network, using deep learning to complicated letter
Several powerful expression abilities, from low-dimensional feature takes out high dimensional feature, represents signal spectrum " portion in overall compact mode
It is point overall " relation, have more essential portraying to target property.
3rd, the fine setting that the present invention carries out having supervision to the network after pre-training using softmax graders is further carried
High discrimination.
Description of the drawings
Fig. 1 is the overall procedure block diagram of the Underwater targets recognition based on deep learning
Fig. 2 is the part diagram of training dataset
Fig. 3 is the part diagram of pretreated training data
Fig. 4 is the elliptical structure figure of stack autoencoder network
Fig. 5 is that corresponding visualization feature exports diagram after 100 hidden nodes of first hidden layer are trained
Fig. 6 is the feature output diagram of a certain node of first hidden layer
Specific embodiment
In conjunction with embodiment, accompanying drawing, the invention will be further described:
Overall procedure based on the Underwater targets recognition of deep learning is as shown in figure 1, comprise the following steps that:
(1) power spectrum characteristic is asked for original Ship Target radiated noise data sectional.
Original Ship Target radiated noise sequence is obtained, the sample frequency of signal is 4KHz, carries out segment processing, is divided
Segment data x (n), n=0,1,2 ..., N-1, according to formulaPower spectrum is carried out to segment data to estimate
Meter, the power spectrum of adjacent 4 segment data is carried out averagely, obtaining final power spectrum characteristic result of calculation:
Wherein f=(f1,f2,f3,…,fM) represent corresponding discrete point in frequency.It is main due to ships radiated noise signal
Spectrum component is concentrated mainly on 0-2KHz, and the points N for calculating FFT is set to the intrinsic dimensionality M of 1024, i.e. one-sided power spectrum herein
For 512, spectral resolution is 4Hz.
(2) power spectrum characteristic is divided into into training dataset and test data set.
By all one-sided power spectrum characteristics according to 3:1 randomly selects, and is divided into training dataset and test data set,
Using each frame power spectrum characteristic of training dataset as string, training matrix P is constituted:
Wherein, Pi(fj) represent that the jth of i-th training data ties up power spectrum characteristic.Accompanying drawing 2 is represented for the part of matrix P.
(3) ZCA whitening pretreatments are carried out to training data;
The first step, is standardized to training data, first calculates the average of kth dimensional feature
Seek the standard deviation of kth dimensional featureCan according to the following formula normalized process after training
Data:
Second step, carries out ZCA whitening pretreatments, first calculates the covariance matrix of sampleSingular value is done to Σ
Characteristic vector is tried to achieve in decomposition, is discharged by leu license-master time order, and is constituted matrix U=[u1,u2,…,un], by eigenvalue by right
Angle arranges, and composition characteristic value diagonal matrix S can be calculated as follows final ZCA albefaction results:
Wherein, ε is used as regularization parameter.Accompanying drawing 3 is through pretreated matrix XZCAPart represent.
(4) construct and train stack autoencoder network.
The first step:Construction stack autoencoder network, constructs stack autoencoder network using sparse self-encoding encoder, and stack is certainly
The hidden layers numbers of coding network are 3, and the interstitial content of hidden layer is respectively 100-100-50, nodes and the ZCA albefactions of input layer
Pretreated training data intrinsic dimensionality is mutually all 512, and finally, increasing by one layer of softmax grader is used for output of classifying, always
Classification number be 3, the nodes of whole network are 512-100-100-50-3.Accompanying drawing 4 shows for the omission of stack autoencoder network
It is intended to.
Second step, using the training data X after ZCA whitening pretreatmentsZCATraining stack autoencoder network, first, according to instruction
The method of the single sparse self-encoding encoder of white silk, using XZCAAs input, the node parameter W of first hidden layer is calculated(1,1)And b(1,1),
Concrete grammar is as follows:
Step 1. is for the input data set X comprising L sampleZCA, define the overall cost function of sparse self-encoding encoder
Wherein, λ is weight attenuation parameter, is set to 3e-3, and β is the parameter for controlling openness penalty factor, and being set to 3, KL is
Relative entropy.
Step 2. is by W(1,1)And b(1,1)Random initializtion is one be close to zero value, using L-BFGS iterative algorithms to generation
Valency function is optimized, and iterationses are set to 100, is met the optimal solution of requirement.Accompanying drawing 5 is 100 of the first hidden layer hidden
After node layer training, corresponding visualization feature output, is divided into 100 row and shows.Accompanying drawing 6 is the single feature of wherein a certain node
Output, " spike " in figure represent node and line-spectrum detection are carried out on different Frequency points.
3rd step, using the activation value of the first hidden layer as the input of the second hidden layer, calculates the second hidden layer with same method
Node parameter W(2,1)And b(2,1), by that analogy, successively train, obtain the parameter of all hidden layers.
Last hidden layer is exported the input as softmax graders, by L-BFGS iterative algorithm iteration by the 4th step
The optimal solution for obtaining softmax graders 100 times.
(5) finely tune depth network model.
Based on back-propagation algorithm, (fine-tune) is finely adjusted to the stack autoencoder network after pre-training, is obtained most
Whole network node parameter.
(6) Classification and Identification is carried out to test data.
According to the variance of the characteristic mean of training dataset, pretreatment test data, using the data for having processed as stack
The input of autoencoder network, calculates final classification results.
The effect of the present invention is illustrated by following experiment embodiment.Experimental data is from certain carried out in South China Sea
Test on secondary sea.The naval vessel of A types, Type B, c-type three types is used respectively as object ship, and hydrophone lays depth for 50
Rice, each type of naval vessel are navigated by water according to left and right by way of in different beam positions, and data acquisition module sample frequency sets
It is set to 4KHz.Table 1 below gives the recognition result that target recognition is carried out by the method for the present invention.As can be seen from Table 1, for
Three kinds of targets, method of the present invention recognition result reach more than 94%.
1 the inventive method recognition result of table
Target recognition is carried out to experimental data using RBF-SVM and BPNN, is contrasted with the method for the present invention, as a result such as
Shown in table 2.As can be seen from Table 2, the discrimination of the method that the present invention is adopted is higher than the RBF-SVM and BPNN of current popular,
Discrimination difference high 4.1% and 5.7%.
2 three kinds of method recognition result contrasts of table
Above test result indicate that, proposed by the invention is effectively may be used based on the Underwater targets recognition of deep learning
Capable, it is possible to increase the identification probability to Ship Target, undersea long is detected and is recognized with important practical significance.
Claims (3)
1. a kind of Underwater targets recognition based on deep learning, it is characterised in that step is as follows:
Step 1:Power spectrum characteristic is asked for original Ship Target radiated noise data sectional:
Obtain original Ship Target radiated noise sequence, carry out segment processing, the segment data for obtaining be x (n), n=0,1,
2 ..., N-1, according to formulaPower Spectral Estimation is carried out to segment data, by adjacent 4 segment data
Power spectrum is carried out averagely, obtains final power spectrum characteristic result of calculation:
Wherein f=(f1,f2,f3,…,fM) represent corresponding discrete point in frequency;
Step 2:Power spectrum characteristic is divided into into training dataset and test data set:
By power spectrum characteristic P (f) according to 3:1 randomly selects, and is divided into training dataset and test data set, by training dataset
Each frame power spectrum characteristic as string, constitute training matrix P:
Wherein, Pi(fj) represent that the jth of i-th training data ties up power spectrum characteristic;
Step 3:ZCA whitening pretreatments are carried out to training data;
Step 4:Construct and train stack autoencoder network;
Step 5:Fine setting fine-tune depth network model:
Using back-propagation algorithm, all layers of stack autoencoder network are regarded into a model, in each iteration, optimize net
All weighted values in network, obtain final network node parameter;
Step 6:Classification and Identification is carried out to test data:
According to the variance of the characteristic mean of training dataset, pretreatment test data will be the data for having processed self-editing as stack
The input of code network, calculates final classification results.
2. a kind of Underwater targets recognition based on deep learning according to claim 1, it is characterised in that step 3 institute
That what is stated carries out ZCA whitening pretreatments to training data and comprises the following steps that:
The first step:Training data is standardized, the average of kth dimensional feature is first calculatedAsk
The standard deviation of kth dimensional featureCan according to the following formula normalized process after training number
According to:
Second step:ZCA whitening pretreatments are carried out, the covariance matrix of sample is first calculatedSingular value decomposition is done to ∑
Characteristic vector is tried to achieve, is discharged by leu license-master time order, is constituted matrix U=[u1,u2,…,un], by eigenvalue by diagonal arrangement,
Composition characteristic value diagonal matrix S, can be calculated as follows final ZCA albefaction results:
Wherein, ε is used as regularization parameter.
3. a kind of Underwater targets recognition based on deep learning according to claim 1, it is characterised in that step 4 institute
The construction stated simultaneously trains comprising the following steps that for stack autoencoder network:
The first step:Construction stack autoencoder network, constructs stack autoencoder network, stack own coding using sparse self-encoding encoder
The hidden layers numbers of network are 3, and the interstitial content of hidden layer is respectively 100-100-50, and nodes and the ZCA albefactions of input layer are located in advance
Training data intrinsic dimensionality after reason is mutually all 512, and finally, increasing by one layer of softmax grader is used for output of classifying, total class
Shuo not be 3, the nodes of whole network are 512-100-100-50-3;
Second step:Using the training data X after ZCA whitening pretreatmentsZCATraining stack autoencoder network, it is first, single according to training
The method of individual sparse self-encoding encoder, using XZCAAs input, the node parameter W of first hidden layer is calculated(1,1)And b(1,1), specifically
Method is as follows:
Step 1) for the input data set X comprising L sampleZCA, define the overall cost function of sparse self-encoding encoder
Wherein, λ is weight attenuation parameter, is set to 3e-3, and β is the parameter for controlling openness penalty factor, and it is relative to be set to 3, KL
Entropy;
Step 2) by W(1,1)And b(1,1)Random initializtion is one be close to zero value, using L-BFGS iterative algorithms to cost function
It is optimized, iterationses are set to 100, is met the optimal solution of requirement;
3rd step:Using the activation value of the first hidden layer as the input of the second hidden layer, the section of the second hidden layer is calculated with same method
Point parameter W(2,1)And b(2,1), by that analogy, successively train, obtain the parameter of all hidden layers;
4th step:Using last hidden layer output as softmax graders input, by L-BFGS iterative algorithms iteration 100 times
Obtain the optimal solution of softmax graders.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610929714.XA CN106529428A (en) | 2016-10-31 | 2016-10-31 | Underwater target recognition method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610929714.XA CN106529428A (en) | 2016-10-31 | 2016-10-31 | Underwater target recognition method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106529428A true CN106529428A (en) | 2017-03-22 |
Family
ID=58292008
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610929714.XA Pending CN106529428A (en) | 2016-10-31 | 2016-10-31 | Underwater target recognition method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106529428A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107315996A (en) * | 2017-05-19 | 2017-11-03 | 燕山大学 | A kind of noise characteristic extracting method of ships under water based on IMF Energy-Entropies and PCA |
CN107798349A (en) * | 2017-11-03 | 2018-03-13 | 合肥工业大学 | A kind of transfer learning method based on the sparse self-editing ink recorder of depth |
CN108387880A (en) * | 2018-01-17 | 2018-08-10 | 西安大衡天成信息科技有限公司 | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes |
CN108599882A (en) * | 2018-03-30 | 2018-09-28 | 中国电子科技集团公司第三十六研究所 | A kind of broader frequency spectrum cognitive method and device based on self-encoding encoder |
CN108694346A (en) * | 2017-04-05 | 2018-10-23 | 中国科学院声学研究所 | A kind of Ship Radiated-Noise signal recognition method based on two-stage CNN |
CN109547374A (en) * | 2018-11-23 | 2019-03-29 | 泰山学院 | A kind of depth residual error network and system for subsurface communication Modulation Identification |
CN109655815A (en) * | 2018-11-23 | 2019-04-19 | 杭州电子科技大学 | Sonar target detection method based on SSD |
CN109932708A (en) * | 2019-03-25 | 2019-06-25 | 西北工业大学 | A method of the underwater surface class object based on interference fringe and deep learning |
CN109961102A (en) * | 2019-03-30 | 2019-07-02 | 北京市商汤科技开发有限公司 | Image processing method, device, electronic equipment and storage medium |
CN110414509A (en) * | 2019-07-25 | 2019-11-05 | 中国电子科技集团公司第五十四研究所 | Stop Ship Detection in harbour based on the segmentation of extra large land and feature pyramid network |
CN112613425A (en) * | 2020-12-24 | 2021-04-06 | 山东船舶技术研究院 | Target identification method and system for small sample underwater image |
CN112946682A (en) * | 2021-01-08 | 2021-06-11 | 浙江大学 | Underwater laser target recognition instrument with high accuracy and high agility |
CN113138365A (en) * | 2020-01-17 | 2021-07-20 | 中国科学院声学研究所 | Single-vector hydrophone direction estimation method based on deep learning |
CN113205050A (en) * | 2021-05-09 | 2021-08-03 | 西北工业大学 | Ship radiation noise line spectrum extraction method based on GRU-AE network |
CN113554123A (en) * | 2021-09-18 | 2021-10-26 | 江苏禹治流域管理技术研究院有限公司 | Automatic sand production ship identification method based on acousto-optic linkage |
GB2607290A (en) * | 2021-05-28 | 2022-12-07 | Bae Systems Plc | Apparatus and method of classification |
CN116953677A (en) * | 2023-09-18 | 2023-10-27 | 海底鹰深海科技股份有限公司 | Sonar target recognition algorithm based on deep learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
CN104156728A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method based on stacked code and softmax |
CN104751173A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Polarized SAR (Synthetic Aperture Radar) image classifying method based on cooperative representation and deep learning. |
CN105787517A (en) * | 2016-03-11 | 2016-07-20 | 西安电子科技大学 | Polarized SAR image classification method base on wavelet sparse auto encoder |
CN105825223A (en) * | 2016-03-09 | 2016-08-03 | 西安电子科技大学 | Polarization SAR terrain classification method based on deep learning and distance metric learning |
-
2016
- 2016-10-31 CN CN201610929714.XA patent/CN106529428A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
CN104156728A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method based on stacked code and softmax |
CN104751173A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Polarized SAR (Synthetic Aperture Radar) image classifying method based on cooperative representation and deep learning. |
CN105825223A (en) * | 2016-03-09 | 2016-08-03 | 西安电子科技大学 | Polarization SAR terrain classification method based on deep learning and distance metric learning |
CN105787517A (en) * | 2016-03-11 | 2016-07-20 | 西安电子科技大学 | Polarized SAR image classification method base on wavelet sparse auto encoder |
Non-Patent Citations (2)
Title |
---|
XU CAO等: ""Deep Learning-Based Recognition of Underwater Target"", 《2016 IEEE INTERNATIONAL CONFERENCE ON DIGITAL SIGNAL PROCESSING》 * |
刘义海等: ""一种优化的舰船基频特征提取算法"", 《鱼雷技术》 * |
Cited By (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108694346B (en) * | 2017-04-05 | 2021-12-07 | 中国科学院声学研究所 | Ship radiation noise signal identification method based on two-stage CNN |
CN108694346A (en) * | 2017-04-05 | 2018-10-23 | 中国科学院声学研究所 | A kind of Ship Radiated-Noise signal recognition method based on two-stage CNN |
CN107315996A (en) * | 2017-05-19 | 2017-11-03 | 燕山大学 | A kind of noise characteristic extracting method of ships under water based on IMF Energy-Entropies and PCA |
CN107798349A (en) * | 2017-11-03 | 2018-03-13 | 合肥工业大学 | A kind of transfer learning method based on the sparse self-editing ink recorder of depth |
CN107798349B (en) * | 2017-11-03 | 2020-07-14 | 合肥工业大学 | Transfer learning method based on depth sparse self-coding machine |
CN108387880B (en) * | 2018-01-17 | 2020-06-09 | 西安大衡天成信息科技有限公司 | Multi-scale directed Lyapunov index-based method for detecting weak and small targets in sea clutter background |
CN108387880A (en) * | 2018-01-17 | 2018-08-10 | 西安大衡天成信息科技有限公司 | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes |
CN108599882B (en) * | 2018-03-30 | 2020-10-27 | 中国电子科技集团公司第三十六研究所 | Self-encoder-based broadband spectrum sensing method and device |
CN108599882A (en) * | 2018-03-30 | 2018-09-28 | 中国电子科技集团公司第三十六研究所 | A kind of broader frequency spectrum cognitive method and device based on self-encoding encoder |
CN109655815A (en) * | 2018-11-23 | 2019-04-19 | 杭州电子科技大学 | Sonar target detection method based on SSD |
CN109547374A (en) * | 2018-11-23 | 2019-03-29 | 泰山学院 | A kind of depth residual error network and system for subsurface communication Modulation Identification |
CN109932708B (en) * | 2019-03-25 | 2022-09-23 | 西北工业大学 | Method for classifying targets on water surface and underwater based on interference fringes and deep learning |
CN109932708A (en) * | 2019-03-25 | 2019-06-25 | 西北工业大学 | A method of the underwater surface class object based on interference fringe and deep learning |
CN109961102A (en) * | 2019-03-30 | 2019-07-02 | 北京市商汤科技开发有限公司 | Image processing method, device, electronic equipment and storage medium |
CN110414509A (en) * | 2019-07-25 | 2019-11-05 | 中国电子科技集团公司第五十四研究所 | Stop Ship Detection in harbour based on the segmentation of extra large land and feature pyramid network |
CN113138365A (en) * | 2020-01-17 | 2021-07-20 | 中国科学院声学研究所 | Single-vector hydrophone direction estimation method based on deep learning |
CN113138365B (en) * | 2020-01-17 | 2022-12-06 | 中国科学院声学研究所 | Single-vector hydrophone direction estimation method based on deep learning |
CN112613425A (en) * | 2020-12-24 | 2021-04-06 | 山东船舶技术研究院 | Target identification method and system for small sample underwater image |
CN112946682A (en) * | 2021-01-08 | 2021-06-11 | 浙江大学 | Underwater laser target recognition instrument with high accuracy and high agility |
CN112946682B (en) * | 2021-01-08 | 2023-03-03 | 浙江大学 | Underwater laser target recognition instrument with high accuracy and high agility |
CN113205050A (en) * | 2021-05-09 | 2021-08-03 | 西北工业大学 | Ship radiation noise line spectrum extraction method based on GRU-AE network |
CN113205050B (en) * | 2021-05-09 | 2023-05-26 | 西北工业大学 | Ship radiation noise line spectrum extraction method based on GRU-AE network |
GB2607290A (en) * | 2021-05-28 | 2022-12-07 | Bae Systems Plc | Apparatus and method of classification |
CN113554123A (en) * | 2021-09-18 | 2021-10-26 | 江苏禹治流域管理技术研究院有限公司 | Automatic sand production ship identification method based on acousto-optic linkage |
CN113554123B (en) * | 2021-09-18 | 2022-01-21 | 江苏禹治流域管理技术研究院有限公司 | Automatic sand production ship identification method based on acousto-optic linkage |
CN116953677A (en) * | 2023-09-18 | 2023-10-27 | 海底鹰深海科技股份有限公司 | Sonar target recognition algorithm based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106529428A (en) | Underwater target recognition method based on deep learning | |
CN112364779B (en) | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion | |
CN109993280B (en) | Underwater sound source positioning method based on deep learning | |
CN109800700B (en) | Underwater acoustic signal target classification and identification method based on deep learning | |
Cao et al. | Deep learning-based recognition of underwater target | |
CN105488466B (en) | A kind of deep-neural-network and Acoustic Object vocal print feature extracting method | |
CN111709315A (en) | Underwater acoustic target radiation noise identification method based on field adaptation | |
CN109034054B (en) | Harmonic multi-label classification method based on LSTM | |
CN113571067B (en) | Voiceprint recognition countermeasure sample generation method based on boundary attack | |
Chen et al. | The research of underwater target recognition method based on deep learning | |
Yang et al. | A new cooperative deep learning method for underwater acoustic target recognition | |
CN110456332A (en) | A kind of underwater sound signal Enhancement Method based on autocoder | |
CN113111786B (en) | Underwater target identification method based on small sample training diagram convolutional network | |
CN112183582A (en) | Multi-feature fusion underwater target identification method | |
CN114897002A (en) | LPINet-based low interception probability radar signal intra-pulse modulation identification method | |
CN107368841A (en) | A kind of underwater manoeuvre Small object recognition methods based on HHT and artificial neural network | |
CN113780521B (en) | Radiation source individual identification method based on deep learning | |
Yu et al. | Underwater target classification using deep neural network | |
CN116430317A (en) | Radiation source modulation pattern and individual identification method and system | |
CN115293214A (en) | Underwater sound target recognition model optimization method based on sample expansion network | |
Xie et al. | DBM-based underwater acoustic source recognition | |
CN114818789A (en) | Ship radiation noise identification method based on data enhancement | |
CN114330450A (en) | Method and system for detecting and identifying underwater vehicle by fusing multiple physical fields | |
Song et al. | A deep-shallow network for passive underwater target recognition | |
Kou et al. | Signals recognition of underwater acoustic communication based on artificial neural network and signal feature extraction |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |